[parentheses]

The timing makes me wonder if this is a direct response to Deepseek V4 having performance comparable to SOTA models.

[osti]

This was published two months ago. Even though it was at a time that open source models are publishing comparable swe bench scores.